Dataset card: English README (primary) + Chinese README_zh, cross-linked
Browse files- README.md +24 -22
- README_zh.md +50 -0
README.md
CHANGED
|
@@ -12,37 +12,39 @@ size_categories:
|
|
| 12 |
- 100K<n<1M
|
| 13 |
---
|
| 14 |
|
| 15 |
-
# Math
|
|
|
|
|
|
|
| 16 |
|
| 17 |
> [!IMPORTANT]
|
| 18 |
-
> **
|
| 19 |
|
| 20 |
-
##
|
| 21 |
|
| 22 |
-
|
| 23 |
|
| 24 |
-
-
|
| 25 |
-
-
|
| 26 |
-
-
|
| 27 |
|
| 28 |
-
##
|
| 29 |
|
| 30 |
-
|
|
| 31 |
|---|---|
|
| 32 |
-
| `preprints/` |
|
| 33 |
-
| `lean/` | Lean
|
| 34 |
-
| `reasoning_traces/` |
|
| 35 |
-
| `overview.pdf` / `overview.tex` |
|
| 36 |
-
| `CONTENTS.md` |
|
| 37 |
-
| `history.md` |
|
| 38 |
|
| 39 |
-
##
|
| 40 |
|
| 41 |
-
-
|
| 42 |
-
-
|
| 43 |
-
-
|
| 44 |
|
| 45 |
-
##
|
| 46 |
|
| 47 |
-
-
|
| 48 |
-
-
|
|
|
|
| 12 |
- 100K<n<1M
|
| 13 |
---
|
| 14 |
|
| 15 |
+
# Math Dataset (Mirror of OpenAI upstream)
|
| 16 |
+
|
| 17 |
+
**English** | [中文说明](./README_zh.md)
|
| 18 |
|
| 19 |
> [!IMPORTANT]
|
| 20 |
+
> **This repository is a mirror, not an original release.** All content originates from OpenAI's upstream repository [openai/math](https://github.com/openai/math). This dataset is **synced with the upstream periodically**, keeping files (including binaries) identical to upstream. For the original English documentation, see the upstream [README](https://github.com/openai/math/blob/main/README.md).
|
| 21 |
|
| 22 |
+
## About this dataset
|
| 23 |
|
| 24 |
+
This dataset contains mathematical manuscripts and supporting proof artifacts produced by an internal OpenAI model. As part of model development, upstream evaluated the model on open research problems and organized the outputs into this collection. This mirror replicates the full collection on Hugging Face for easy download and research use.
|
| 25 |
|
| 26 |
+
- The catalogue contains **719 manuscripts** organized into **372 families** (a family groups related papers: a principal result, companion arguments, consequences, or alternative proofs), classified by mathematical discipline.
|
| 27 |
+
- Results are at different stages of verification; not all have accompanying Lean formalizations. The upstream continues to add formalizations — currently about **42% of top-line results are formalized**.
|
| 28 |
+
- Some unformalized results could have issues; upstream will fix them quickly.
|
| 29 |
|
| 30 |
+
## Repository structure
|
| 31 |
|
| 32 |
+
| Path | Description |
|
| 33 |
|---|---|
|
| 34 |
+
| `preprints/` | Paper PDFs, LaTeX sources, per-paper citation and build instructions |
|
| 35 |
+
| `lean/` | Lean formal proof library, including the `formalization.yaml` catalogue and verification configs |
|
| 36 |
+
| `reasoning_traces/` | Abridged summaries of the model's reasoning (10 PDFs) |
|
| 37 |
+
| `overview.pdf` / `overview.tex` | Descriptions of the families |
|
| 38 |
+
| `CONTENTS.md` | Manuscript map for finding individual papers |
|
| 39 |
+
| `history.md` | Upstream update history |
|
| 40 |
|
| 41 |
+
## Sync policy
|
| 42 |
|
| 43 |
+
- This repository is periodically synced from upstream `openai/math`, aiming to match the latest upstream commit exactly.
|
| 44 |
+
- Syncs are performed by [@roseking](https://huggingface.co/roseking). If anything is missing, feel free to open an Issue.
|
| 45 |
+
- For details on how the results were produced, versioning, and citation, refer to the upstream repository.
|
| 46 |
|
| 47 |
+
## License and citation
|
| 48 |
|
| 49 |
+
- Upstream content is released under the **Apache License 2.0** (see `LICENSE`); all rights remain with the original authors. This mirror follows the same license.
|
| 50 |
+
- To cite an individual manuscript, use the BibTeX block in its directory, and prefer citing the original upstream repository.
|
README_zh.md
ADDED
|
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
- zh
|
| 6 |
+
tags:
|
| 7 |
+
- mathematics
|
| 8 |
+
- mirror
|
| 9 |
+
- lean
|
| 10 |
+
- formal-proofs
|
| 11 |
+
size_categories:
|
| 12 |
+
- 100K<n<1M
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# Math 数据集(OpenAI 上游镜像)
|
| 16 |
+
|
| 17 |
+
[English](./README.md) | **中文说明**
|
| 18 |
+
|
| 19 |
+
> [!IMPORTANT]
|
| 20 |
+
> **本仓库是镜像(mirror),不是原创发布。** 所有内容的原始出处为 OpenAI 的上游仓库 [openai/math](https://github.com/openai/math)。本数据集会**定期与上游同步**,文件(含二进制文件)保持与上游一致。原始英文说明请以上游 [README](https://github.com/openai/math/blob/main/README.md) 为准。
|
| 21 |
+
|
| 22 |
+
## 数据集简介
|
| 23 |
+
|
| 24 |
+
本数据集包含由 OpenAI 内部模型产出的数学论文手稿及其配套证明材料。上游在模型开发过程中于公开研究问题上评估模型能力,并将产出整理为此合集。本镜像将其完整同步至 Hugging Face,便于下载与研究使用。
|
| 25 |
+
|
| 26 |
+
- 当前目录包含 **719 篇手稿**,组织为 **372 个论文家族**(family:同一主结果的相关论文、推论或替代证明的分组),按数学学科分类。
|
| 27 |
+
- 各结果处于不同的验证阶段,并非全部附带 Lean 形式化;上游会持续更新补充,目前约 **42% 的主要结果已完成形式化**。
|
| 28 |
+
- 部分未经形式化的结果可能存在问题,上游会尽快修正。
|
| 29 |
+
|
| 30 |
+
## 目录结构
|
| 31 |
+
|
| 32 |
+
| 路径 | 说明 |
|
| 33 |
+
|---|---|
|
| 34 |
+
| `preprints/` | 论文 PDF、LaTeX 源文件、各论文的引用与构建说明 |
|
| 35 |
+
| `lean/` | Lean 形式化证明库,含 `formalization.yaml` 目录与验证配置 |
|
| 36 |
+
| `reasoning_traces/` | 模型推理过程的节选摘要(10 篇 PDF) |
|
| 37 |
+
| `overview.pdf` / `overview.tex` | 各论文家族的综述介绍 |
|
| 38 |
+
| `CONTENTS.md` | 手稿索引(快速定位单篇论文) |
|
| 39 |
+
| `history.md` | 上游更新历史 |
|
| 40 |
+
|
| 41 |
+
## 同步说明
|
| 42 |
+
|
| 43 |
+
- 本仓库定期从上游 `openai/math` 同步,目标与上游最新提交保持一致。
|
| 44 |
+
- 同步由维护者 [@roseking](https://huggingface.co/roseking) 执行;如发现缺漏,欢迎提 Issue。
|
| 45 |
+
- 结果的生成方式、版本管理与引用规范等细节,请以上游仓库说明为准。
|
| 46 |
+
|
| 47 |
+
## 许可与引用
|
| 48 |
+
|
| 49 |
+
- 上游内容以 **Apache License 2.0** 发布(见本仓库 `LICENSE`),版权归原作者所有;本镜像同样遵循该许可。
|
| 50 |
+
- 引用单篇论文时,请使用其目录下提供的 BibTeX 条目,并优先引用上游原始仓库。
|